high performance computing (hpc) on gacrc sapelo2...
TRANSCRIPT
High Performance Computing (HPC) on GACRC Sapelo2 Cluster
Georgia Advanced Computing Resource Center
EITS/University of Georgia
Zhuofei Hou, [email protected]
1HIGH PERFORMANCE COMPUTING (HPC) ON CLUSTER10/16/2019
Outline
• GACRC
• What is High Performance Computing (HPC)?
• What is a Cluster?
• How to Work on Cluster?
2HIGH PERFORMANCE COMPUTING (HPC) ON CLUSTER10/16/2019
Wiki: http://wiki.gacrc.uga.edu
Support: https://uga.teamdynamix.com/TDClient/Requests/ServiceCatalog?CategoryID=11593
Web Site: http://gacrc.uga.edu
3HIGH PERFORMANCE COMPUTING (HPC) ON CLUSTER10/16/2019
GACRC A high-performance-computing (HPC) center at the UGA
Provide to the UGA research and education community an advanced computing
environment:
• HPC computing and networking infrastructure located at the Boyd Data Center
• Comprehensive collection of scientific, engineering and business applications
• Consulting and training services
What is High Performance Computing?
4HIGH PERFORMANCE COMPUTING (HPC) ON CLUSTER10/16/2019
TaskMountain
1 worker
10 years later……
What is High Performance Computing? (cont.)
5HIGH PERFORMANCE COMPUTING (HPC) ON CLUSTER10/16/2019
TaskMountain
2 months later……
What is High Performance Computing? (cont.)
Serial Jobwith 1 worker
Parallel Jobwith 6 workers
Only 1 instruction executed at any time point on a single processor
6 instructions executed simultaneously
on 6 processors
Shared-memory threaded parallel job or
MPI distributed parallel job (Message
Passing Interface)
610/16/2019 HIGH PERFORMANCE COMPUTING (HPC) ON CLUSTER
What is a Cluster?
When you work on cluster, 3 roles are working:
You: Who submit a job
Queueing System: Who dispatch your job to run on cluster
Cluster: Who run your job
7HIGH PERFORMANCE COMPUTING (HPC) ON CLUSTER10/16/2019
Submit a job Dispatch a job
Please Note: You need to connect to the UGA network using VPN when accessing from outside of the UGA main campus.UGA VPN: https://eits.uga.edu/access_and_security/infosec/tools/vpn/
8HIGH PERFORMANCE COMPUTING (HPC) ON CLUSTER10/16/2019
Two Nodes:
1. Login node for batch job workflow: [email protected]
2. Transfer node for data transferring: [email protected]
Five Directories:
1. Home: Login landing spot; 100GB quota; Backed-up
2. Scratch: High-speed storage for temp files needed for current jobs; NO quota; NOT backed-up
3. Work: High-speed storage for input files needed for repeated jobs; per group quota of 500GB and max
100,000 single files; NOT backed-up
4. Project: Temporary data parking; per group quota of 1TB; Backed-up (ONLY accessible from Transfer node!)
5. Local Scratch: Local storage on each individual compute node; 200GB quota; NOT backed-up
Four Computational Queues: batch, highmem_q, gpu_q, groupBuyin_q
9HIGH PERFORMANCE COMPUTING (HPC) ON CLUSTER10/16/2019
Overview https://wiki.gacrc.uga.edu/wiki/Systems#Sapelo2
Overview (cont.)
To connect to cluster, you need to have a user account:
Once your account is provisioned, use SSH Secure Shell program to open connection: https://wiki.gacrc.uga.edu/wiki/Connecting
10HIGH PERFORMANCE COMPUTING (HPC) ON CLUSTER10/16/2019
1. Group PI request a user account for you: http://help.gacrc.uga.edu/userAcct.php
2. You need to pass new user training: https://wiki.gacrc.uga.edu/wiki/Training
3. We send you a welcome letter with detailed info about your Sapelo2 user account
How to work on cluster?
11HIGH PERFORMANCE COMPUTING (HPC) ON CLUSTER10/16/2019
Your job working space is global scratch folder /scratch/MyID/
• Why?
High-performance filesystem with the fastest I/O!
• How to access?
From Login node, cd /scratch/MyID
• What do you need?
Software! We installed ~1000 software on cluster for you. Check if the software
you need is already installed at https://wiki.gacrc.uga.edu/wiki/Software
• What do you need to pay attention to?
Clean up data you will not need after your job is finished!
lustre1
lustre1
Workflow Diagram
12
1. Linux/Mac user:ssh [email protected]
Windows user:
5. Linux/Mac user:scp file [email protected]:/scratch/MyID/workDir
Windows user:
Login6. nano ./sub.sh
$ qsub sub.sh
#PBS -S /bin/bash#PBS -q batch#PBS -N bowtie2_test#PBS -l nodes=1:ppn=1:Intel#PBS -l mem=2gb#PBS -l walltime=1:00:00
#PBS -M [email protected]#PBS -m ae
cd $PBS_O_WORKDIR
module load Bowtie2/2.3.3-foss-2016b……
2. cd /scratch/MyID
3. mkdir ./workDir
4. cd ./workDir
7.
HIGH PERFORMANCE COMPUTING (HPC) ON CLUSTER10/16/2019
$ qstat_me or
qdel JobID8.
Useful Links• GACRC Web: http://gacrc.uga.edu/
• GACRC Wiki: https://wiki.gacrc.uga.edu/wiki/Main_Page
• GACRC FAQ: https://wiki.gacrc.uga.edu/wiki/Sapelo2_Frequently_Asked_Questions
• GACRC Help: http://gacrc.uga.edu/help/
• GACRC Training: https://wiki.gacrc.uga.edu/wiki/Training
• GACRC User Account: https://wiki.gacrc.uga.edu/wiki/User_Accounts
• GACRC Software: https://wiki.gacrc.uga.edu/wiki/Software
HIGH PERFORMANCE COMPUTING (HPC) ON CLUSTER10/16/2019 13
Thank You!
14HIGH PERFORMANCE COMPUTING (HPC) ON CLUSTER10/16/2019